Edtech Development at Startup House: What We've Built and What It Changed

Key facts at a glance

• Flagship EdTech build: Graspify, a microlearning platform for corporate education, taken from concept through MVP to version 2.0, with version 2.1 planned, deployed in Norwegian corporations
• Proof of scale: in August 2022, a Graspify client used the platform to train hundreds of employees during a corporate event
• Discovery-first practice: LITTLEWINE, a B2B wine education platform, scoped through 10 user interviews in 2 weeks with winemakers, buyers, sommeliers, and consumers
• AI learning tools: KnowHub and InProduct AI can go live in as little as 2 weeks once content is prepared
• Delivery model: phased builds with separate budgets per stage and real user testing before every next phase

Why EdTech products fail, and what we do differently

EdTech platforms fail for UX reasons more often than technical ones. Learners drop off when interfaces don't match how people actually learn. Content doesn't scale. Onboarding takes too long. ROI stays invisible to the people paying for the platform.

Our answer is process discipline: define the outcome first, build in stages with individual budgets per phase, and test with real learners before each next step. Here is how that played out in the projects below.

Graspify: from a bold idea to live corporate training

The client: Graspify, a Norwegian startup out to change corporate employee education with a microlearning platform combining learning with a research-based learning process. Cooperation from January 2021.

The challenge: Turn a bold vision into a functional product, and find the right form for "Grasps", the short 5-6 slide mini-courses at the heart of the platform. How long should one be? How complex? How simple? That balance was the core design problem, and it could not be solved by opinion.

How we worked: We split the project into smaller steps, each with its own budget and its own round of user testing, both 1-on-1 and in friendly groups. In January 2021 we completed Product Discovery and the vision for Graspify 1.0. Then came the MVP (versions 1.0 and 1.1), followed by the extended version 2.0, with version 2.1 planned on the basis of live usage data.

What changed:

• Graspify was deployed in several Norwegian corporations.
• In August 2022, a Graspify client used MVP 2.0 to train hundreds of employees during a corporate event.
• The events generated significant data for analysis, feeding directly into the plan for version 2.1.

The platform lets companies create custom content, invite employees, manage access, track progress, and recommend courses. The phased model meant Graspify never committed a full build budget to an unvalidated assumption.

Read the Graspify case study

LITTLEWINE: scoping a knowledge platform with the people who will use it

The client: LITTLEWINE, a British B2B SaaS democratizing wine education and knowledge exchange between professionals and winemakers. Cooperation from November 2021.

The challenge: Make wine knowledge available directly from winemakers, without intermediaries diluting authenticity, and validate the B2B direction before committing to development.

How we worked: A series of workshops covering competitor analysis, customer needs, and potential revenue streams, then wireframe prototyping grounded in user knowledge and market research. The key move: 10 user interviews across 2 groups (winemakers, buyers, sommeliers, consumers) conducted within 2 weeks.

What changed: LITTLEWINE received a complete B2B product scope definition and MVP estimates in several variants, built for pitching to early investors. Features included dedicated winemaker accounts with content upload, personalized collections, a database of regions and grape varieties, and advanced search.

This is what EdTech discovery looks like when the audience is niche: you talk to the actual professionals, quickly, before anyone writes production code.

Read the LITTLEWINE case study

AI in EdTech: learning tools that only answer from your content

For learning platforms, a wrong answer is not a bug. It undermines learning outcomes and can create liability. That is why all three of our AI products for EdTech operate on a verified-answers model: the AI responds only using approved content, cannot generate answers from public training data, and cites every response back to a source document.

• KnowHub turns your knowledge base into a 24/7 learning assistant, answering in any language, on any device.
• InProduct AI reads your codebase, writes its own documentation, and guides users through your platform and courses.
• SmartSearch understands what a learner is looking for, not just the words they typed, and verifies each result with a proof of fact.
• For existing platforms, we also build a conversational AI layer on top of legacy LMS systems, so users get natural-language access to data without a system rewrite.

KnowHub and InProduct AI can deploy in as little as 2 weeks once your content or codebase is prepared. Custom platform builds follow a phased model, typically 4 to 12 weeks to a first usable version depending on scope.

What we bring beyond AI

Full-stack EdTech delivery covers the rest of what a learning product needs:

• UX and learning experience design: onboarding flows, course structures, habit-forming progress tracking, and assessment UX, user-tested at every stage.
• Analytics and learning intelligence: dashboards showing what content works, where learners drop off, and what predicts completion before a learner churns.
• Cloud infrastructure: built on GCP, AWS, or Azure to handle traffic spikes, concurrent sessions, and global users, with CI/CD and 24/7 support tiers.
• Platform integrations: SSO, HRIS, CRM, payment providers, and content libraries, so the platform fits into any enterprise stack.

Data governance holds across all of it: learner data is never used to train shared or public AI models, under the same ISO 27001-compliant practices we apply for enterprise clients like Siemens Healthineers.

What this changed, in one paragraph

For Graspify, phased delivery turned a bold vision into a product running live corporate training and generating its own improvement data. For LITTLEWINE, two weeks of structured interviews turned a broad idea into an investor-ready scope. The common thread: in EdTech, evidence about learners beats assumptions about learners, at every stage of the build.

About Startup House

Startup House is a 50-person software development company based in Warsaw, Poland, established in 2016, with 100+ products shipped for clients on 5 continents. We work with early-stage EdTech startups and established learning platforms alike, from discovery through MVP to long-term iteration.

Building an EdTech product? Book a 30-min call.

FAQ

What EdTech products has Startup House built?

Startup House built Graspify, a corporate microlearning platform deployed in Norwegian corporations and used to train hundreds of employees during a corporate event, and scoped LITTLEWINE, a B2B wine education platform, through structured discovery with 10 user interviews in 2 weeks. We also deploy AI learning tools (KnowHub, InProduct AI, SmartSearch) for existing platforms.

Does Startup House work with early-stage EdTech startups?

Yes. Graspify started as a concept with no product, and we delivered discovery, design, the MVP, and version 2.0. We also work with established platforms adding AI capabilities or scaling infrastructure.

How long does it take to build an EdTech platform?

AI learning tools like KnowHub and InProduct AI can go live in as little as 2 weeks once content is prepared. Custom builds follow a phased model, typically 4 to 12 weeks to a first usable version depending on scope, with user testing built into each phase.

How does Startup House make sure AI gives accurate answers in a learning context?

All our AI products for EdTech operate on a verified-answers model: the AI responds only from approved content, cannot answer from public training data, and cites every response back to a specific source document.

Is learner data used to train AI models?

Never. Learner data stays within the client's environment and is not used to train shared or public AI models, under ISO 27001-compliant data governance.